Favard Length Problem

• Favard length problem is the study of the average length of orthogonal projections of planar sets, quantifying decay rates for self-similar fractals.
• It employs geometric, arithmetic, and Fourier analytic methods to derive precise lower and upper bounds on projection lengths.
• The topic bridges classical geometric measure theory with modern techniques including nonlinear projections and randomized fractal models.

The Favard length problem concerns the quantitative behavior of the average length of orthogonal projections of planar sets. For a planar set $E\subseteq \mathbb{R}^2$, the Favard length is defined by

$$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$

where $\pi_\theta$ denotes orthogonal projection onto the line in direction $\theta$. It is also interpreted as Buffon needle probability, and it is closely related to rectifiability, Hausdorff dimension, and analytic capacity (Bongers, 2017). In its classical form, the problem asks how fast $\operatorname{Fav}(E_n)$ or $\operatorname{Fav}(E(r))$ decays for approximating generations $E_n$ of self-similar planar Cantor sets, or for $r$-neighborhoods $E(r)$, when the limiting set is purely $1$-unrectifiable.

1. Classical formulation and qualitative theory

The qualitative background is the Besicovitch projection theorem: if $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$0 has $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$1 and is purely $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$2-unrectifiable, then $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$3. In this sense, Favard length is a projection-theoretic detector of rectifiable structure (Dąbrowski, 2024). For self-similar approximants $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$4, however, the limit statement $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$5 leaves open the central quantitative question: how fast the decay occurs (Laba, 2012).

The canonical examples are dimension-$$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$6 planar Cantor sets such as the four-corner set, the modified Sierpiński gasket, and rational product sets. For these, Mattila’s lower bound gives

$$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$7

while for the four-corner set Bateman and Volberg improved the lower bound to

$$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$8

The exact decay rate remains unresolved in the deterministic self-similar setting, and this unresolved rate is the core of the classical Favard length problem for planar Cantor sets (Laba, 2012).

A parallel formulation uses neighborhoods. If $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$9 is a compact purely $\pi_\theta$0-unrectifiable set with $\pi_\theta$1, then the problem becomes the asymptotic behavior of $\pi_\theta$2 as $\pi_\theta$3. This neighborhood version is particularly useful when one wants to compare deterministic, random, and nonlinear projection models.

2. Deterministic self-similar sets and arithmetic methods

For deterministic self-similar sets, the best upper bounds are highly sensitive to algebraic structure. Łaba’s survey records power-law bounds

$\pi_\theta$4

for several canonical families, including the four-corner set, the Sierpiński gasket, self-similar sets with $\pi_\theta$5, and rational product sets satisfying a tiling condition. It also records the weaker bound

$\pi_\theta$6

for all rational product sets with $\pi_\theta$7, and the much weaker general estimate

$\pi_\theta$8

for general self-similar sets (Laba, 2012).

The paper "The Favard length of product Cantor sets" generalizes the Nazarov–Peres–Volberg framework from the four-corner set to product Cantor sets whose projection in some direction has positive $\pi_\theta$9-dimensional measure. If $\theta$0 in lowest terms and $\theta$1 has positive measure, then there exists $\theta$2 such that

$\theta$3

where $\theta$4 for the two factor dimensions (Laba et al., 2009). The decisive input is the arithmetic structure of exceptional projections, mediated by tiling theory and generating-function identities.

In the rational product setting, the harmonic-analytic reduction passes through mask polynomials and cyclotomic divisibility. Łaba emphasizes the generating polynomials

$\theta$5

the trigonometric polynomial

$\theta$6

and the Set of Small Values (SSV) and Set of Large Values (SLV) mechanisms (Laba, 2012). The paper "Vanishing sums of roots of unity and the Favard length of self-similar product sets" sharpens the Lam–Leung lower bound on vanishing sums of roots of unity and extends the Bond–Łaba–Volberg method to sets for which the least common multiples $\theta$7 and $\theta$8 of the relevant cyclotomic divisors each have at most two distinct prime divisors. In that regime one obtains

$\theta$9

and if all roots of $\operatorname{Fav}(E_n)$0 and $\operatorname{Fav}(E_n)$1 on the unit circle are roots of unity, the bound improves to

$\operatorname{Fav}(E_n)$2

The same work raises the size threshold handled by this method from $\operatorname{Fav}(E_n)$3 to $\operatorname{Fav}(E_n)$4 (Laba et al., 2022).

These results show that the deterministic problem is not governed by self-similarity alone. Fourier decay, cyclotomic factorization, and tiling phenomena enter at full strength, and the boundary between power-law decay and weaker decay is strongly arithmetic.

3. Geometric bounds, convexity, and limits of self-similar heuristics

A distinct line of work replaces Fourier analysis by direct geometry. "Geometric Bounds for Favard Length" proves that if $\operatorname{Fav}(E_n)$5 is measurable, $\operatorname{Fav}(E_n)$6 has positive measure, $\operatorname{Fav}(E_n)$7, and

$\operatorname{Fav}(E_n)$8

for some $\operatorname{Fav}(E_n)$9, then

$\operatorname{Fav}(E(r))$0

The proof is geometric and uses coverings and Hölder’s inequality rather than Frostman measures or energy integrals (Bongers, 2017).

The same paper proves a convexity property for generations of self-similar sets. If $\operatorname{Fav}(E(r))$1 with $\operatorname{Fav}(E(r))$2 and $\operatorname{Fav}(E(r))$3, then for each direction $\operatorname{Fav}(E(r))$4,

$\operatorname{Fav}(E(r))$5

is convex in $\operatorname{Fav}(E(r))$6, in the sense that

$\operatorname{Fav}(E(r))$7

This yields lower bounds on Favard length for several self-similar fractals; for the four-corner Cantor generations it gives

$\operatorname{Fav}(E(r))$8

by a purely geometric argument (Bongers, 2017).

The self-similar picture has strict limitations. "Sets with Arbitrarily Slow Favard Length Decay" constructs measurable purely unrectifiable sets $\operatorname{Fav}(E(r))$9 with $E_n$0 such that for any increasing sequence $E_n$1,

$E_n$2

Equivalently, for any monotone $E_n$3 with $E_n$4, there exists such an $E_n$5 with

$E_n$6

for all small $E_n$7 (Wilson, 2017). This shows that the familiar logarithmic and power-law patterns from self-similar Cantor sets do not extend to arbitrary purely unrectifiable $E_n$8-sets.

A common misconception is that self-similar lower bounds reflect a universal phenomenon. The non-self-similar constructions show that no universal lower decay law of Mattila type can hold in that generality.

4. Rectifiability, quantitative rigidity, and large Favard length

Another major theme asks what large Favard length forces geometrically. "Structure of sets with nearly maximal Favard length" studies finite-length sets $E_n$9 against a line segment $r$0 with $r$1. Since

$r$2

line segments maximize Favard length among sets of a given length. If

$r$3

then $r$4 can be covered by an $r$5-Lipschitz graph up to a set of length $r$6, with polynomial dependence

$r$7

Thus near-maximizers are quantitatively close to straight line segments (Chang et al., 2022).

The Ahlfors-regular theory goes further. "Favard length and quantitative rectifiability" proves a quantitative Besicovitch theorem: if $r$8 is Ahlfors $r$9-regular with constant $E(r)$0 and

$E(r)$1

then there exists a Lipschitz graph $E(r)$2 with $E(r)$3 such that

$E(r)$4

Moreover, for Ahlfors regular sets, the condition

$E(r)$5

is equivalent to having Big Pieces of Lipschitz Graphs and hence to uniform rectifiability (Dąbrowski, 2024).

That work also connects Favard length to analytic capacity. For Ahlfors regular sets with uniformly large Favard length, it proves a quantitative lower bound of analytic capacity in terms of diameter, providing a finite-length Ahlfors-regular case of Vitushkin’s conjecture (Dąbrowski, 2024). At the opposite end, it gives an explicit general upper bound for Ahlfors regular purely unrectifiable sets: $E(r)$6 where $E(r)$7 is defined by supremizing $E(r)$8 over curves $E(r)$9 (Dąbrowski, 2024).

These results reposition Favard length as a quantitative rectifiability functional, not merely a decay observable for fractals.

5. Nonlinear projections and generalized Favard functionals

The Favard length framework extends beyond orthogonal projections. "Transversal families of nonlinear projections and generalizations of Favard length" introduces a general transversality condition for nonlinear projection-type families $1$0. If the family is $1$1-transversal and $1$2 supports a probability measure with Frostman growth $1$3, then for the $1$4-neighborhood $1$5,

$1$6

and

$1$7

This yields nonlinear analogues of Mattila-type lower bounds for visibility, Favard curve length, and Favard surface length (Bongers et al., 2021).

For the four-corner Cantor set, the same framework gives

$1$8

for piecewise $1$9 curves with piecewise bi-Lipschitz continuous unit tangent vectors (Bongers et al., 2021). "Upper and lower bounds on the rate of decay of the Favard curve length for the four-corner Cantor set" establishes matching nonlinear upper and lower bounds for a large class of curves $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$00: $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$01 and

$$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$02

In that setting the key ingredients are a local comparison between curve projections and orthogonal projections, and a counting-function $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$03 method for the lower bound (Cladek et al., 2020).

The nonlinear Besicovitch theorem has also been quantified. "A Quantification of a Besicovitch Nonlinear Projection Theorem via Multiscale Analysis" proves that if $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$04 has controlled multiscale length and sufficiently small quantitative rectifiability constants $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$05, then for suitable piecewise $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$06 curves $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$07,

$$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$08

As an application, for the four-corner Cantor set one obtains the upper bound

$$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$09

and, together with the companion work of Cladek, Davey, and Taylor, a power-law upper bound $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$10 for all $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$11 (Davey et al., 2021).

This nonlinear theory shows that the classical Favard length problem is part of a broader projection program in which orthogonal projections, curve projections, visibility, and surface probes are handled by a common transversality-and-energy mechanism.

6. Random models, higher dimensions, and extremal problems

Random models often exhibit much sharper decay than deterministic self-similar sets. "The exact Power Law for Buffon's needle landing near some Random Cantor Sets" proves that for random $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$12-adic disk Cantor sets generated by independent random rotations,

$$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$13

and similarly for random $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$14-adic models with $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$15. Together with Mattila’s lower bound, this gives the exact $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$16 decay rate in the average sense (Zhang, 2018). "The Buffon's needle problem for random planar disk-like Cantor sets" proves the same order

$$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$17

for a third random disk model, matching the lower bound and reinforcing the observation that several distinct randomization schemes yield the same logarithmic law (Vardakis et al., 2022).

More recent work shows that this law is neither accidental nor universal. "Sharp Favard length of random Cantor sets" proves that for a large class of planar $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$18-dimensional random fractals $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$19,

$$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$20

with almost sure asymptotics and an explicit limiting constant for wide classes of grid random fractals. The same paper also constructs $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$21-dimensional Ahlfors-regular random fractals for which

$$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$22

showing that $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$23 is not universal even among random Ahlfors-regular sets (Chang et al., 19 Dec 2025).

The scope of the problem has also expanded to higher dimensions. "Power Laws for the Favard Length Problem in $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$24" proves power-law upper bounds for neighborhoods of higher-dimensional analogues of the four-corner Cantor set and for broader rational digit constructions. For the $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$25-corner Cantor set $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$26,

$$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$27

and for $$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$28 this is stated to be the first non-trivial asymptotic upper bound for the Favard length problem in that setting (Marshall, 2 Sep 2025).

A separate extremal direction treats Favard length as a variational functional. "The isoperimetric problem for the Favard length" shows that among planar Borel sets of fixed area, a circle minimizes Favard length. For the unit-area disk,

$$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$29

and more generally

$$\operatorname{Fav}(E)=\int_0^{\pi} |\pi_\theta(E)|\,d\theta,$$30

with equality only for circles in the stated sense (Karasev et al., 9 Jun 2026). Together with the near-maximizer theory for sets of fixed length, this places Favard length within a broader extremal geometry.

The contemporary Favard length problem is therefore no longer a single decay estimate for one Cantor set. It is a network of quantitative questions about projection size, arithmetic self-similarity, rectifiability, randomness, nonlinear probing, and extremal geometry. The deterministic planar four-corner problem remains the emblematic test case, but the surrounding theory now reaches from cyclotomic divisibility to analytic capacity and from Ahlfors-regular quantitative rectifiability to higher-dimensional self-similar constructions.